Estimating soil thermal diffusivity at different water contents from easily available data on soil texture, bulk density, and organic carbon content

This study provides an algorithm to estimate soil thermal diffusivity at any water content from data on soil texture, bulk density, and percentage of organic carbon. Models were trained on the dataset of 77 soil samples including silty clays, silty clay loams, silt loams, clay loams, loams, sandy clay loams, sandy loams, loamy sands, and sands. The ranges of sand, silt, and clay within the dataset were 1-97, 2-80, and 1-52%; wet bulk density varied from 860 to 1820 kg m(-3), organic carbon ranged from 0.1 to 6.5%. Thermal diffusivity of the undisturbed soil cores measured by the unsteady-state method was from 0.77 to 10.09 x 10(-7) m(2) s(-1). The dataset was split randomly into the training set of 67 samples and the test set of 10 samples; the procedure was repeated three times. Models were developed from the measured thermal diffusivity vs. water content curves. The experimental data points for each sample were described by a 4-parameter function. Parameters of average curves for different textural classes were also determined. Then regression equations were obtained to estimate the parameters of the thermal diffusivity vs. water content function for different soils: (i) from soil texture; (ii) from soil texture and bulk density; (iii) from soil texture and organic carbon; (iv) from soil texture, bulk density, and organic carbon. The test set data were used to evaluate the model performance. The normalised root mean square errors of the best-performing models were from 20 to 33% depending on soil information available. (C) 2017 IAgrE. Published by Elsevier Ltd. All rights reserved.